Let $(\Omega,\mathcal{A},\mu)$ be a $\sigma$-finite measure space and $\lambda$ be the Lebesgue-measure on $\mathbb{R}$. Moreover let $f\geq 0$ be a measurable numerical function on $\Omega$. Show, that for the set $$ A_f:=\left\{(\omega,x)\in\Omega\times\mathbb{R}: 0\leq x\leq f(\omega)\right\} $$ it is $$ (\mu\otimes\lambda)(A_f)=\int_{\Omega}f\, d\mu=\int_0^{\infty}\mu(\left\{f\geq x\right\})\, d\lambda(x). $$ Hint for the proof: First show the assumption for elementar functions and approximate any non-negative measurable function by elementar functions from below.


Following the given hint, I first tried to show it for an (non-negative) elementar function, i.e. $f\colon\Omega\to [0,\infty)$ with $f:=\sum_{i=1}^{n}\alpha_i 1_{A_i}$, whereat $A_i, i=1,\ldots,n\in\mathcal{A}$ pairwise disjoint and $\alpha_i\geq 0$.

$$ (\mu\otimes\lambda)(A_f)=\int_{\Omega}\int_{\mathbb{R}}(1_{A_f})_{\omega}(x)\, d\lambda(x)\, d\mu(\omega)=\int_{\Omega}\int_0^{f(\omega)}\, d\lambda(x)\, d\mu(\omega)=\int_{\Omega}\lambda([0,f(\omega)])\, d\mu(\omega)\\=\int_{\Omega}f(\omega)\, d\mu(\omega)=\sum_{i=1}^{n}\alpha_i\mu(A_i) $$

Now I start from the RHS: $$ \left\{ f\geq x\right\}=\biguplus_{i=1}^{n}\left\{f=\alpha_i: \alpha_i\geq x\right\} $$ $$ \implies\int_0^{\infty}\mu(\left\{f\geq x\right\})\, d\lambda(x)=\int_0^{\infty}\sum_{i=1}^{n}\mu(\left\{f=\alpha_i: \alpha_i\geq x\right\})\, d\lambda(x)\\=\sum_{i=1}^{n}\int_{0}^{\infty}\mu(\left\{f=\alpha_i:\alpha_i\geq x\right\})\, d\lambda(x)\\=\sum_{i=1}^{n}\int_0^{\infty}\int_{\Omega}1_{\left\{f=\alpha_i: \alpha_i\geq x\right\}}(\omega)\, d\mu(\omega)\, d\lambda(x)\\=\sum_{i=1}^{n}\int_0^{\infty}\int_{\Omega}1_{\left\{f=\alpha_i\right\}}(\omega)1_{x\leq\alpha_i}(x)\, d\mu(\omega)\, d\lambda(x)\\=\sum_{i=1}^{n}\int_0^{\infty}1_{x\leq\alpha_i}(x)\underbrace{\int_{\Omega}1_{\left\{ f=\alpha_i\right\}}(\omega)}_{=\mu(A_i)}\, d\mu(\omega)\, d\lambda(x)\\=\sum_{i=1}^{n}\mu(A_i)\int_0^{\infty}1_{x\leq\alpha_i}(x)\, d\lambda(x)\\=\sum_{i=1}^{n}\mu(A_i)\underbrace{\int_0^{\alpha_i}\, d\lambda(x)}_{=\lambda([0,\alpha_i])=\alpha_i}\\=\sum_{i=1}^{n}\alpha_i\mu(A_i) $$

So the assumption is shown for an (non-negative) elementar function.

Now let $f$ be a non-negative, numerical, measurable function. Then there exists a sequence $(f_n)$ of non-negative elementar functions with $f_n\uparrow f$.

Because $A_{f_n}\subset A_{f_{n+1}}$ and $A_{f_n}\uparrow A_f$ it is with the continuity of the measure and Beppo Levi $$ (\mu\otimes\lambda)(A_f)=\lim_{n\to\infty}(\mu\otimes\lambda)(A_{f_n})=\lim_{n\to\infty}\int_{\Omega}f_n\, d\mu=\int_{\Omega}f\, d\mu. $$

Furthermore by Beppo Levi it is

$$ \int_{\Omega}f\, d\mu=\lim_{n\to\infty}\int_{\Omega}f_n\, d\mu\\=\lim_{n\to\infty}\int_{0}^{\infty}\mu(\left\{f_n\geq x\right\})\, d\lambda(x) $$

Now the only thing that remains to show is that $$ \lim_{n\to\infty}\int_{0}^{\infty}\mu(\left\{f_n\geq x\right\})\, d\lambda(x)=\int_0^{\infty}\mu(\left\{f\geq x\right\})\, d\lambda(x) $$

Unfortunately, I do not see that.

Can you explain that to me?

With greetings!



Since $f_n\uparrow f$, then $\{f_n\geq x\}\uparrow \{f\geq x\}$ and hence also $\mu(\{f_n\geq x\})\uparrow \mu(\{f\geq x\})$, so the claim follows from the monotone convergence theorem.

To show the measurability of $x\mapsto \mu(\{g\geq x\})$ for measurable $g:\Omega\to\mathbb{R}$, we need the following result:

Let $(X,\mathcal{E},\mu)$ and $(Y,\mathcal{F},\nu)$ be two $\sigma$-finite measure spaces. Let also $U\in\mathcal{E}\otimes\mathcal{F}$ and define the sets $$ U_x=\{y\in Y\mid (x,y)\in U\},\quad U^y=\{x\in X\mid (x,y)\in U\}. $$ Then $U_x\in\mathcal{F}$, $U^y\in\mathcal{E}$ and if we define the mappings $\varphi_U: X\to [0,\infty)$ and $\psi_U:Y\to [0,\infty)$ by $$ \varphi_U(x)=\nu(U_x),\quad \psi_U(y)=\mu(U^y), $$ then $\varphi_U$ is $\mathcal{E}$-$\mathcal{B}(\mathbb{R})$-measurable and $\psi_U$ is $\mathcal{F}$-$\mathcal{B}(\mathbb{R})$-measurable

If we define $H$ by $$ H=\{(\omega,x)\in\Omega\times \mathbb{R}\mid 0\leq x\leq g(\omega)\} $$ then $$ H=p_2^{-1}([0,\infty))\cap(f\circ p_1-p_2)^{-1}([0,\infty))\in \mathcal{A}\otimes\mathcal{B}(\mathbb{R}) $$ where $p_1$ and $p_2$ are the projections on $\Omega\times \mathbb{R}$ (which are measurable). Now, $$ H^x=\{\omega\in\Omega\mid (\omega,x)\in H\}= \begin{cases} \varnothing,\quad &\text{if }x<0,\\ \{g\geq x\},\quad &\text{if }x\geq 0, \end{cases} $$ and by the result above we get that $$ x\mapsto \mu(\{f\geq x\})=\mu(H^x) $$ is Borel-measurable.

  • $\begingroup$ For using the monotone convergence theorem, it is necessary, that the functions $x\mapsto\mu(\left\{f_n\geq x\right\})$ are measurable, isn't it? And I do not see that. $\endgroup$ – math12 Jan 26 '14 at 10:37
  • $\begingroup$ Let $[a,b)$ be an interval in $[0,\infty]$ and consider the preimage $P$ of it under the mapping $x\mapsto\mu(\{f_n\geq x\})$. $P$ is the set of such $x\in[0,\infty)$ that $a\leq\mu(\{f_n\geq x\})<b$. Since $\mu(\{f_n\geq x\})$ is decreasing in $x$, $P$ must be an interval as well (possibly an infinite one), which is measurable. $\endgroup$ – triple_sec Jan 26 '14 at 10:50
  • $\begingroup$ Before I try to understand your proof, one further question. In order to show that $x\mapsto \mu(\left\{f_n\geq x\right\})=\int_{\Omega}1_{\left\{f_n\geq x\right\}}(\omega)\, d\mu(\omega)$ is measurable, I thought I have to show that the function $1_{\left\{f_n\geq x\right\}}\colon \Omega\times\mathbb{R}\to\mathbb{R}$ is measurable, because then Fubini implies that $x\mapsto\mu(\left\{f_n\geq x\right\})$ is measurable? But this is, what I am not able to show. $\endgroup$ – math12 Jan 26 '14 at 10:58
  • $\begingroup$ Fubini works, too, but it may be more cumbersome to check whether $1_{\{f_n\geq x\}}$ is measurable in the product measure. $\endgroup$ – triple_sec Jan 26 '14 at 11:02
  • 1
    $\begingroup$ @math12 Provided you can see why the monotonicity of $x\mapsto\mu(\{f_n\geq x\})$ implies that $\{x\in\mathbb R\,|\,a<\mu(\{f_n\geq x\}<b\}$ is an interval, note that any interval (open, closed, half-open, half-closed, finite, or infinite) can be generated by open sets through countably many operations of taking unions, intersections, and complements. But $\mathcal B(\mathbb R)$, the Borel $\sigma$-algebra generated by open sets, is, by the very definition, closed under these operations, so that the resulting interval is a Borel set. $\endgroup$ – triple_sec Jan 26 '14 at 18:32

Another way of showing the result, without taking the route via simple functions, is to use Tonelli's theorem (which applies thanks to $\sigma$-finiteness and positivity of the integrand) and write $$\int_{0}^{\infty}\mu\left\{ f\geq x\right\} d\lambda=\int\int\chi_{\left\{ 0<x\leq f\right\} }d\mu d\lambda\overset{\text{Tonelli}}{=}\int\int\chi_{\left\{ 0<x\leq f\right\} }d\lambda d\mu=\int\int_{0}^{f}d\lambda d\mu=\int fd\mu.$$

This shows equality of the integrals, and note that the second term is exactly the product measure of $A_f$, again by Tonelli's theorem.

Edit: Note that as pointed out in the comments by @triple_sec, this takes for granted the measurability of the involved functions.

  • $\begingroup$ But you would still need to show the joint $\mu\otimes\lambda$-measurability of $(\omega,x)\mapsto\chi_{\{x\leq f(\omega)\}}$; that is, the $\mathcal A\otimes\mathcal B$-measurability of the set $$\{(\omega,x)\in\Omega\times\mathbb R\,|\,f(\omega)\geq x\},$$ which I guess was something the OP would rather have avoided. $\endgroup$ – triple_sec Jan 26 '14 at 21:15
  • $\begingroup$ OK, that was silly. You have the show the measurability of that set anyway as a part of the exercise! $\endgroup$ – triple_sec Jan 26 '14 at 21:25
  • $\begingroup$ Yeah, but anyway, you're right as to where the bulk of the work lies. $\endgroup$ – ekvall Jan 27 '14 at 9:06

$\textbf{Claim 1:}\quad$ The map $x\mapsto\mu(\{\omega\in\Omega\,|\,h(\omega)\geq x\})$ from $\mathbb R$ to $\overline{\mathbb R}$ is non-increasing for any measurable function $h:\Omega\to\mathbb[0,\infty]$.

$\textit{Proof:}\quad$ Clearly, if $x,x'\in\mathbb R$ and $x<x'$, then $$\{\omega\in\Omega\,|\,h(\omega)\geq x'\}\subseteq \{\omega\in\Omega\,|\,h(\omega)\geq x\},$$ since the set on the right-hand side has a laxer lower bound. Now the claim follows from the monotonicity of measures. (Note that both sets are measurable, since they are upper contour sets of $h$, which is a measurable function.) $\blacksquare$

$\textbf{Claim 2:}\quad$ If $g:\mathbb R\to\overline{\mathbb R}$ is a non-increasing function, and $a,b\in[-\infty,\infty]$ with $a<b$, then the set $$G\equiv\{x\in\mathbb R\,|\,g(x)\in(a,b)\}$$ is (i) empty, (ii) a singleton, or (iii) an interval.

$\textit{Proof:}\quad$ Let $$G_b\equiv\{x\in\mathbb R\,|\,g(x)<b\}$$ and $$G_a\equiv\{x\in\mathbb R\,|\,g(x)>a\}.$$ Clearly, $G=G_a\cap G_b$. Let's focus on $G_b$ first. Let $$x_b\equiv\sup\{x\in\mathbb R\,|\,g(x)\geq b\}.$$ By convention, if this set is empty (implying that $G_b=\mathbb R$), then $x_b=-\infty$. If, on the other hand, $\{x\in\mathbb R\,|\,g(x)\geq b\}$ is non-empty, then we will show that either $G_b=(x_b,\infty)$ or $G_b=[x_b,\infty)$.

Small digression: we can assume that $x_b$ is finite without loss of generality. Clearly, if the set $\{x\in\mathbb R\,|\,g(x)\geq b\}$ is non-empty, then $x_b>-\infty$. On the other hand, if $x_b=\infty$, then $g(x)\geq b$ for all $x\in\mathbb R$, since $g$ is non-increasing (if it went below $b$ at some point, it would stay below $b$ thereafter, which would imply that $x_b<\infty$). In this case, $G_b$ is empty and so is $G$, which would immediately complete the proof. So, from now on, focus on the cases in which $x_b\in(-\infty,\infty)$.

There are two cases:

  • Case 1: $g(x_b)\geq b$. In this case, if $g(x)<b$, then $x>x_b$ (since $g$ is non-increasing), implying that $G_b\subseteq (x_b,\infty)$. Conversely, if $x>x_b$, then $g(x)<b$, since $g(x)\geq b$ would contradict the supremum property of $x_b$. This implies that $x\in G_b$. Hence, $(x_b,\infty)\subseteq G_b$. Conclusion: $G_b=(x_b,\infty)$.

  • Case 2: $g(x_b)<b$. Now, suppose that $x\in G_b$. If $x<x_b$ (for the sake of contradiction), then, by the supremum property of $x_b$, there exists some $x'\in(x,x_b]$ such that $g(x')\geq b$. That is, $x<x'$, but (since $x\in G_b$) $g(x)<b\leq g(x')$, contradicting $g$ being non-increasing. Hence, we have that $x\geq x_b$. This implies that $G_b\subseteq [x_b,\infty)$. Next, if $x\geq x_b$, then $g(x)\leq g(x_b)<b$, so that $x\in G_b$. Conclusion: $G_b=[x_b,\infty)$.

Hence, $$G_b\in\{\mathbb R,(x_b,\infty),[x_b,\infty)\},$$ and, by the same token, $$G_a\in\{\mathbb R,(-\infty,x_a),(-\infty,x_a]\},$$ where $$x_a\equiv\inf\{x\in\mathbb R\,|\,g(x)\leq a\}.$$ Again, there is no loss of generality in assuming that $x_a$ is finite.

Now, $G=G_a\cap G_b$ can take one of following forms: $$\mathbb R,(x_b,\infty),[x_b,\infty),(-\infty,x_a),(-\infty,x_a],(x_b,x_a),(x_b,x_a],[x_b,x_a),[x_b,x_a].$$ The first five of these are unambiguously infinite intervals. The last four sets are empty if $x_b>x_a$. (In fact, one can prove that $x_b\leq x_a$ holds, but we don't need this for the desired result.) If $x_b=x_a$, then the last one is a singleton, while $(x_b,x_a)$, $(x_b,x_a]$, and $[x_b,x_a)$ are still empty. Finally, if $x_b<x_a$, then the last four are non-degenerate finite intervals. In particular, $G$ is measurable in all cases. $\blacksquare$


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.